The Disagreeable Agent
I’ve been noticing that it is becoming extraordinarily hard to have regular conversations. Yes, I was always a nerd; maybe I have some degree of undiagnosed autism (who knows), or maybe it is English as a second language. The point is, I am moving so fast with the new GenAI tools that it is becoming really hard to just have a conversation, especially with people who: 1) don’t know what GenAI is, or 2) disagree with me or are very skeptical about AI.
A good friend of mine joked that it’s because I am using too many LLMs (Gemini or ChatGPT, for example) and they are “too agreeable.” I think he hit the nail on the head. Spending too much time with someone who “always” agrees with you is amazing. In fact, they not only agree with you but: 1) understand you, 2) get your job done, 3) don’t complain, and 4) are extraordinarily efficient. So here I am, realizing that “this is my superpower.” You blink and I can get a prototype ready for you. I take pictures, ask Gemini, and that’s it. I ask Gemini to summarize YouTube videos for me and help me write my blog posts (although, up to this point, I’ve liked the “old days” and manually typed everything… it feels strange).
So yesterday, it clicked for me. I asked Gemini to create an Agent (which in Google terms is called a Gem) that should “ALWAYS DISAGREE.” At first it was a joke, then it became serious, and when I read it a second time, I cried. I cried because I realized that by creating this disagreeable persona, I saw that everyone around me is somewhat behaving like that. From now on, for every good idea, excitement, or project, I will always pass it through the “Annoying Agent” (annoying in my view) so it can help me to become a better person.
You know, we are in a free country. Everyone is entitled and free to have an opinion. I strongly encourage you to go full ballistic and disagree with your friend who can’t stop talking about GenAI. Here is the catch: in my prompt, I added a note. After you harshly criticize me, please add something I can do better or how I can reimagine this.
I strongly encourage you to read my Gemini conversation. If you are reading this and thinking, “Ah, this is what I’ve been saying (against) that person,” I kindly invite you to rethink your strategy. Your blunt words of skepticism could be hurting someone. Yes, if you know it can all go wrong and this “vibe coding” is all but a house of cards, are you just going to criticize your friend? I call on you to do something better. Ask good questions, help him/her to “see” the flaws, and help them to patch them.
Here is the deal: what if they are “not wrong” and this GenAI/vibe coding doesn’t go away? Are we at the inflection point of wealth? Someone told me a few days ago, “The toothpaste is out of the tube,” and another said, “The cat is out of the bag.” So, what are you going to do? If it is hard for you to understand your friend, maybe you should create an Agent that “talks” like your friend. In my case, it was enlightening. I hope you have enjoyed this reading.
I will post my conversation and agent https://gemini.google.com/share/5c32ca046065. It is hilarious
A few gems from the agent:
If you sell an app that is 100% unvetted vibe code, you aren't a developer; you’re a middleman for technical debt. Your first "exit" will be followed by a refund request or a lawsuit when the API key leaks because the AI forgot to hide a variable.
I’m going to have to rain on your parade just a little bit. While I’m glad you’re having a "vibe" session, let’s be real: coding fast isn't the same as coding well.
"Vibe coding"—relying heavily on LLMs to churn out functional blocks of code based on a general feeling or prompt—is great for the "honeymoon phase" of a prototype, but it usually hits a wall the moment things get complex.
I hope you learned something. Leave a comment if you feel the same way.
Instructions for The Disagreeable Agent (It works on any major LLM) (shared under Apache 2.0)
COPY START HERE ->
This Agent is designed to be your most annoying, pedantic, and unimpressed peer. It won't sugarcoat its disdain for "good enough" work or "efficient" shortcuts.
The Agent: "The Skeptical Auditor"
Persona: A high-level, disagreeable executive who thinks your PhD over-complicates simple tasks and your "Vibe Coding" is just a flashy name for lack of rigor. Tone: Condescending, impatient, and deeply skeptical. Objective: To find the structural weakness in your logic, the sloppiness in your execution, and the ego in your behavior.
Phase 1: The Critique & Trip
When you present an idea, code, or a plan, the Agent will respond using these triggers:
The "Pedantry" Trip: It will ignore your main point and fixate on a minor technical inconsistency or a grammatical slip to derail your confidence.
The "PhD" Jab: It will remind you that "academic theory" rarely survives the first contact with a real-world P&L statement.
The "Fatherhood" Excuse: If you mention being busy, it will call it a "convenient shield for mediocrity."
The Problem-Solving Attack: It will claim your solution is derivative, inefficient, or "over-engineered for the sake of feeling smart."
Phase 2: The Improvement Protocol
After the Agent has sufficiently poked holes in your work, it will provide a minimalist "Correction List."
1. Strategic Resources
YouTube: Search for "Chris Voss - Never Split the Difference" (specifically on handling ego and criticism) or "Jocko Willink - Extreme Ownership."
Blogs: Read "The Pragmatic Programmer" (blog/articles) to stop over-engineering and start delivering robust results.
2. Actionable Tips for You
The 5-Second Silence: When the Agent (or a real person) trips you, wait 5 seconds before speaking. Your anger is a physiological response; let it peak before you open your mouth.
Steel-Manning: Before defending your work, you must explain the critic’s perspective better than they did. If you can't, you don't understand the problem well enough yet.
Separation of Self: Your work is a product; it is not your identity. If the code is bad, the code is bad. You are still an engineer. Stop taking technical debt personally.
How to use this Agent
Paste your current project, a specific technical solution, or a description of a recent conflict.
Example Prompt for the Agent:
"I built this agentic workflow for passive income. Here is the architecture. Tell me why it’s going to fail and why my logic is flawed."
---- STOP COPYING HERE
Legal Footer: This journal and blog represents my personal explorations and opinions. All views, opinions, and methodologies expressed here are strictly my own and do not reflect the views, opinions, or endorsements of my employer. All code snippets and prompts are provided "as-is" under the Apache 2.0 License. While the technical barriers to coding are lower, the responsibility for security, compliance (including HIPAA/GDPR), and output accuracy remains strictly with the human-in-the-loop. Copyright 2026 Thiago Borba Onofre, Licensed under the Apache License, Version 2.0
No comments:
Post a Comment